The Macroeconomics of Epidemics: Martin S. Eichenbaum

The Macroeconomics of Epidemics
Martin S. Eichenbaum
Northwestern University and NBER
Sergio Rebelo
Northwestern University, NBER, and CEPR
Mathias Trabandt
Downloaded from https://academic.oup.com/rfs/article/34/11/5149/6213914 by guest on 10 November 2021

Goethe University Frankfurt and IWH
We extend the canonical epidemiology model to study the interaction between economic
decisions and epidemics. Our model implies that people cut back on consumption and
work to reduce the chances of being infected. These decisions reduce the severity of the
epidemic but exacerbate the size of the associated recession. The competitive equilibrium
is not socially optimal because infected people do not fully internalize the effect of their
economic decisions on the spread of the virus. In our benchmark model, the best simple
containment policy increases the severity of the recession but saves roughly half a million
lives in the United States. (JEL E1, I1, H0)
Received December 19, 2020; editorial decision February 12, 2021 by Editor Ralph Koijen.
As COVID-19 spreads throughout the world, governments are struggling with

how to understand and manage the epidemic. Epidemiology models have been
widely used to predict the course of the epidemic (e.g., Ferguson et al. 2020).
While these models are very useful, they do have a significant shortcoming:
they do not allow for the interaction between economic decisions and rates of
infection.
Policy makers certainly appreciate this interaction. For example, in an op-ed
piece, Ben Bernanke and Janet Yellen (2020) write: “In the near term, public
health objectives necessitate people staying home from shopping and work,
especially if they are sick or at risk. So production and spending must inevitably
decline for a time.”
We are grateful to Andy Atkeson, Gadi Barlevy, Francisco Ciocchini, Warren Cornwall, Ana Cusolito, João
Guerreiro, Ravi Jaganathan, Aart Kraay, Michael King, Per Krusell, Chuck Manski, Paul Romer, Alp Simsek,
and Steve Strongin for their comments. We thank Bence Bardoczy, Joao Guerreiro, and Laura Murphy for
excellent research assistance. Matlab replication codes can be downloaded from the authors’ websites or directly
from https://tinyurl.com/ERTcode. Send correspondence to Mathias Trabandt, mathias.trabandt@gmail.com.
The Review of Financial Studies 34 (2021) 5149–5187
© The Author(s) 2021. Published by Oxford University Press.
This is an Open Access article distributed under the terms of the Creative Commons Attribution License
(http://creativecommons.org/licenses/by/4.0/), which permits unrestricted reuse, distribution, and reproduction
in any medium, provided the original work is properly cited.
doi:10.1093/rfs/hhab040 Advance Access publication August 7, 2021
[22:50 7/10/2021 RFS-OP-REVF210044.tex] Page: 5149 5149–5187

The Review of Financial Studies / v 34 n 11 2021
In this paper, we extend the canonical SIR model proposed by Kermack and
McKendrick (1927) to study the equilibrium interaction between economic
decisions and epidemic dynamics.1 Our model features a two-way interaction
between the epidemic and the economy. People’s decisions to cut back on
consumption and work reduce the severity of the epidemic, as measured by
total deaths. These same decisions exacerbate the size of the recession caused
by the epidemic.
In our model, an epidemic has both aggregate demand and aggregate supply
effects. The supply effect arises because the epidemic exposes workers to the
virus. Workers react to that risk by reducing their labor supply. The demand
effect arises because the epidemic exposes consumers to the virus. Consumers

react to that risk by reducing consumption. The supply and demand effects
work together to generate a large, persistent recession.
The competitive equilibrium is not Pareto optimal because people infected
with the virus do not fully internalize the effect of their consumption and work
decisions on the spread of the virus. To be clear, this market failure does not
reflect a lack of good intentions or irrationality on the part of infected people. It
simply reflects the fact that each infected person takes economywide infection
rates as given. But collectively, their behavior does change infection rates,
thereby imposing unpriced costs on susceptible people.2
A natural question is: what policies should the government pursue to deal
with the infection externality? We focus on simple containment policies that
reduce consumption and hours worked. By reducing economic interactions
among people, these policies exacerbate the recession but raise welfare by
reducing the death toll caused by the epidemic. We find that it is optimal to
introduce large-scale containment measures that result in a sharp, sustained
drop in aggregate output. In our benchmark model, when health care capacity
is limited and vaccines and treatments don’t arrive before the epidemic is over,
containment policy saves roughly half a million lives in the United States.
To make the intuition for our results as transparent as possible, we use
a relatively simple model. A cost of that simplicity is that we cannot study
many important policy issues related to the epidemic. For example, we do not
consider policies that mitigate the economic hardships suffered by households
and businesses. Such policies include fiscal transfers to households and loans
to keep firms from going bankrupt. We also do not study policies aimed at
maintaining the liquidity and health of financial markets.
Finally, we abstract from nominal rigidities that could play an important
role in determining the short-run response of the economy to an epidemic. For
1 SIR is an acronym for susceptible, infected, recovered, and removed.
2 The behavior of susceptible people is also different in the competitive equilibrium and the Pareto optimum. This
difference can also be interpreted as an externality because it influences the dynamics of the epidemic and thus
affects the number of people killed by the virus. See Rachel (2020) and Garibaldi, Espen, and Pissarides (2020)
for a discussion of this effect.
5150

example, if prices are sticky, a given fall in the demand for consumption would
generate a larger recession. Other things equal, a larger recession would mitigate
the spread of the infection.3 But we are confident that the central message from
our current analysis will be robust: there is an inevitable trade-off between the
severity of the recession and the health consequences of the epidemic.4
Our point of departure is the canonical SIR model proposed by Kermack
and McKendrick (1927). In this model, the transition probabilities between
health states are exogenous parameters. We modify the model by assuming
that purchasing consumption goods and working brings people in contact with
each other. These activities raise the probability that the infection spreads. We
refer to the resultant framework as the SIR-macro model.

We choose parameters so that the Kermack-McKendrick SIR model is
consistent with the scenario outlined by Chancellor Angela Merkel in her speech
on March 11, 2020 (Bennhold and Eddy 2020). According to this scenario,
“60% to 70% of the population will be infected as long as this remains the
situation.” Using 60% as our benchmark value, the SIR model implies that the
share of the initial population infected peaks at 6.8%. Applying this scenario to
the United States implies that roughly 200 million Americans will eventually
become infected and 1 million people will die. An obvious shortcoming of
the SIR model is that people do not take any actions to reduce the chances of
becoming infected and infection dynamics are not influenced by the level of
economic activity.
The interaction between economic activity and transition probabilities in
the SIR-macro model substantially changes the dynamics of the epidemic
and its economic impact. One way to assess this impact is to focus on the
simplest version of the SIR-macro model that abstracts from the possibility of
vaccinations, medical treatments, and limited health care capacity. Relative to
the SIR model, this simple SIR-macro model implies a sharper recession and
fewer deaths. The average fall in aggregate consumption in the first year of the
epidemic is roughly seven times larger than in the SIR model (4.7% vs. 0.7%).
This larger decline in economic activity reduces the infection peak (5.2% vs.
6.8%) as well as the percentage of the population who becomes infected (54%
vs. 60%). Critically, the total number of U.S. deaths caused by the epidemic
falls from 1 million to 890,000.
To design optimal policy, one must understand how epidemics end. In both
the SIR and SIR-macro models, epidemics end when a sufficiently high fraction
of the population acquires immunity so that the number of infections no longer
rises (i.e., the population achieves “herd immunity”). Absent vaccines, the only
way to acquire immunity is to become infected and recover. Sadly, without
effective medical treatments, this process involves the death of many people.
3 In a follow-up to this paper, Eichenbaum, Rebelo, and Trabandt (2020b) incorporate nominal rigidities and
physical investment into the model.
4 In an interesting essay, Gourinchas (2020) makes a similar point.
5151

In all versions of our model, it is optimal for policy makers to avoid recurrent
epidemics. So a key question for policy is: what is the optimal way to reach
herd immunity?
In the SIR-macro model, it is possible to prevent the infection from spreading
by adopting large, permanent containment measures. This approach has two
problems. First, the permanent containment measures create a persistent
economic depression. Second, the population never reaches herd immunity.
So, infections would recur if containment was ever relaxed.
The best policy in this world is to curtail consumption when externalities
are large, that is, when the number of infected people is high. Such a policy
involves gradually ramping up containment measures as infections rise and

slowly relaxing them as new infections wane and the population approaches
herd immunity.
An important concern in many countries is that the health care system can be
overwhelmed by a large number of infected people. To analyze this scenario, we
extend the simple SIR-macro model so that the case fatality rate (the probability
of dying conditional on being infected) is an increasing function of the number
of people infected. We find that the competitive equilibrium involves a much
larger recession as people internalize the higher case fatality rates. People cut
back more aggressively on consumption and work to reduce the probability
of being infected. As a result, fewer people are infected in the competitive
equilibrium, but more people die. The optimal policy involves a much more
aggressive response than in the simple SIR-macro economy. The reason is that
the cost of the externality is much larger since a larger fraction of the infected
population dies.
How does the possibility of an effective treatment being discovered change
our results? The qualitative implications are clear: people become more willing
to engage in market activities because the expected cost of being infected is
smaller. So, along a path in which treatment is not actually discovered, the
recession induced by the epidemic is less severe. Sadly, along such a path, the
total number of infected people and the death toll rise relative to the baseline
SIR-macro model. That said, the quantitative difference between this model
and the baseline SIR-macro model is quite small, with respect to both the
competitive equilibrium and the best containment policy.
How does the possibility of a vaccine being discovered change our results?
Vaccines don’t cure infected people, but they do prevent susceptible people
from becoming infected. In contrast, treatments cure infected people but do not
prevent future infections. Given our benchmark calibration, these differences
are not very important for the competitive equilibrium. But they are very
important for the design of optimal policy. With vaccination as a possibility, it
is optimal to immediately introduce severe containment measures to minimize
deaths. Those measures cause a large recession. But this recession is worth
incurring in the hope that a vaccine arrives before many people get infected.
5152

The most general version of our model, discussed in Section 6, incorporates

the probabilistic development of vaccines and treatments, as well as a case
fatality rate that rises with the number of infected people. The latter feature
reflects capacity constraints in the health care system. We refer to this version
of the model as the benchmark SIR-macro model.
In this model, it is optimal to immediately introduce severe containment
measures and increase those measures as more of the population is infected. The
best containment policy dramatically increases the magnitude of the recession.
Absent containment measures, average consumption falls by about 7% in the
first year of the epidemic. With optimal containment, average consumption
falls by 22%. Notably, the size of the recession is smaller than in the medical

preparedness model. The reason is that the prospect of vaccinations and
treatments reduces the magnitude of the externality associated with the medical
preparedness problem.
The benefit of the large recession associated with optimal containment in
the benchmark model is a less severe epidemic. Compared to the competitive
equilibrium, the peak infection rate drops from 4.7% to 2.5% of the initial
population. The optimal policy reduces the death toll as a percentage of the
initial population from 0.40% to 0.26%. For the United States, this reduction
amounts to about half a million lives.
We emphasize that these numbers pertain to a worst-case scenario in which
vaccines and treatments never arrive. If they do arrive, many more lives would be
saved. Thankfully, they would be saved by medicine rather than by containment
policies.
Finally, we quantify the effects of delaying or prematurely ending optimal
containment policies. Abandoning containment policies prematurely leads to
an initial economic recovery. But it also leads to a large rise in infection rates.
That rise causes a new, persistent recession. Tragically, the overall death toll
rises because optimal policy was abandoned.
Suppose that containment policies are designed and implemented well into
an infection episode. At that point, it is optimal to adopt extreme containment
measures that cause a large recession. The reason is simple: the longer is the
delay, the larger is the number of infections and the externalities associated
with economic activity. Optimal policy then involves draconian containment to
offset those externalities. Even so, the overall death toll is much larger than if
containment had been implemented without delay.
The simple containment strategy that we study mimics a key feature of
existing policies: containment applies equally to everyone, regardless of their
health status. A natural question is: how much better could a benevolent
government do if it could directly choose the consumption and hours worked
of susceptible, infected, and recovered people?
We answer this question by solving the relevant social-planning problem.
This solution, which we call “smart containment,” requires that infected people
don’t work unless they recover. This isolation policy means that susceptible
5153

people can work without the risk of becoming infected. The amount that
susceptible and recovered people consume is the same as in the pre-epidemic
steady state. Consumption of infected people depends on whether it is feasible
to deliver goods to them without the risk of infecting other people. In any
event, the economy does not suffer in any meaningful way from a recession.
Moreover, the overall death toll of the epidemic is very small, with the number
of infected people declining monotonically from its initial level to zero.
The previous results point to the importance of antigen and antibody tests
that would allow health care professionals to quickly ascertain people’s health
status. The social returns to gathering this information and acting on it are
enormous. These actions reduce both the death toll and the size of the economic

contraction relative to the outcomes associated with the best simple containment
policy.
1. The SIR-Macro Model

In this section, we describe the economy before the start of the epidemic. We
then present the SIR-macro model.
1.1 The preinfection economy

The economy is populated by a continuum of ex ante identical people with
measure one. Prior to the start of the epidemic, everybody is identical and
maximizes the objective function
∞

U= β t u(ct ,nt ).
t=0
Here, β ∈ (0,1) denotes the discount factor and ct and nt denote consumption
and hours worked, respectively. For simplicity, we assume that momentary
utility takes the form
θ
u(ct ,nt ) = lnct − n2t .
2
The budget constraint of the representative person is
(1+μt )ct = wt nt +t .
Here, wt denotes the real wage rate, μt is a Pigouvian tax rate on

consumption, and t denotes lump-sum transfers from the government. As
discussed below, we think of μt as a proxy for containment measures aimed at
reducing social interactions. For this reason, we refer to μt as the containment
rate. In Section 5, we study an alternative way to model containment that does
not involve taxation but yields very similar results.
The first-order condition for the representative person’s problem is
(1+μt )θnt = ct−1 wt .
5154

There is a continuum of competitive representative firms of unit measure

that produce consumption goods (Ct ) using hours worked (Nt ) according to
the technology
Ct = ANt .
The firm chooses hours worked to maximize its time-t profits t
t = ANt −wt Nt .
The government’s budget constraint is given by
μt ct = t .

In equilibrium, nt = Nt and ct = Ct .
1.2 The outbreak of an epidemic

Epidemiology models generally assume that the probabilities governing the
transition between different states of health are exogenous with respect to
economic decisions. We modify the canonical SIR model proposed by Kermack
and McKendrick (1927) so that these transition probabilities depend on people’s
economic decisions. Since purchasing consumption goods or working brings
people into contact with each other, we assume that the probability of becoming
infected depends on these activities.
The population is divided into four groups: susceptible (people who have
not yet been exposed to the disease), infected (people who contracted the
disease), recovered (people who survived the disease and acquired immunity),
and deceased (people who died from the disease). The fractions of people in
these four groups are denoted by St , It , Rt , and Dt , respectively. The number
of newly infected people is denoted by Tt .5
Susceptible people can become infected in three ways. First, they can meet
infected people while purchasing consumption goods. The number ofnewly
infected people that results from these interactions is given by π1 (St Cts ) It Cti .
The terms St Cts and It Cti represent total consumption expenditures by
susceptible and infected people, respectively. The parameter π1 reflects both the
amount of time spent shopping and the probability of becoming infected as a
result of that activity. In reality, different types of consumption involve different
amounts of contact with other people. For example, attending a rock concert is
much more contact intensive than going to a grocery store. For simplicity, we
abstract from this type of heterogeneity.
Second, susceptible and infected people can meet at work. The number
of newly infected people that results from interactions at work is given by
π2 (St Nts ) It Nti . The terms St Nts and It Nti represent total hours worked by
5 We assume that people know their current health status. In subsequent work (Eichenbaum, Rebelo, and Trabandt
2020a), we develop a model in which people do not know their health status unless they are tested.
5155

susceptible and infected people, respectively. The parameter π2 reflects the

probability of becoming infected as a result of work interactions. We recognize
that different jobs involve different levels of social contact. For example,
working as a dentist or a waiter is much more contact intensive than writing
software. Again, for simplicity, we abstract from this source of heterogeneity.
Third, susceptible and infected people can meet in ways not directly related
to consuming or working, for example, meeting a neighbor or riding an elevator.
The number of random meetings between infected and susceptible people is
St It . These meetings result in π3 St It newly infected people.
The total number of newly infected people is given by6

Tt = π1 St Cts It Cti +π2 St Nts It Nti +π3 St It . (1)

The Kermack and McKendrick (1927) SIR model is a special case of our
model in which the propagation of the disease is unrelated to economic activity
(π1 = 0, π2 = 0).
The number of susceptible people at time t +1 is equal to the number of
susceptible people at time t minus the number of susceptible people who got
infected at time t:
St+1 = St −Tt . (2)
The number of infected people at time t +1 is equal to the number of infected
people at time t plus the number of newly infected (Tt ) minus the number of
infected people who recovered (πr It ) and the number of infected people who
died (πd It ):
It+1 = It +Tt −(πr +πd )It . (3)
Here, πr is the rate at which infected people recover from the infection and πd
is the case fatality rate, that is, the probability that an infected person dies.
The timing convention implicit in Equation (3) is as follows. Social
interactions happen in the beginning of the period (infected and susceptible
people meet). Then, changes in health status unrelated to social interactions
(recovery or death) occur. At the end of the period, the consequences of social
interactions materialize: Tt susceptible people become infected.
The number of recovered people at time t +1 is the number of recovered
people at time t plus the number of infected people who just recovered (πr It ),
Rt+1 = Rt +πr It . (4)
Finally, the number of deceased people at time t +1 is the number of deceased
people at time t plus the number of new deaths (πd It ),
Dt+1 = Dt +πd It . (5)
6 To simplify, we assume that the probability of a given person being infected through more than one form of
social interactions is zero. In addition, we do not explicitly incorporate the constraint that Tt must be between
zero and the size of population. This constraint is satisfied in all our simulations.
5156

Total population, Popt+1 , evolves according to

Popt+1 = Popt −πd It ,
with Pop0 = 1.
We assume that at time zero, a fraction ε of susceptible people is infected
by a virus through zoonotic exposure, that is, the virus is directly transmitted
from animals to humans:
I0 = ε,
S0 = 1−ε.
Everybody is aware of the initial infection and understands the laws of

motion governing population health dynamics. Critically, people take as given
aggregate variables, such as It Cti and It Nti .
We now describe the optimization problem of different types of people in
j
the economy. The variable Ut denotes the time-t lifetime utility of a type-j
person (j = s,i,r). The budget constraint of a type-j person is
j j
(1+μt )ct = wt φ j nt +t , (6)
j j
where ct and nt denote the consumption and hours worked of a type-j person,
respectively. The parameter governing labor productivity, φ j , is equal to one
for susceptible and recovered people (φ s = φ r = 1) and less than one for infected
people (φ i < 1).
The budget constraint (6) embodies the assumption that people cannot pool
risk associated with the infection. Going to the opposite extreme and assuming
complete markets considerably complicates the analysis without necessarily
making the model more realistic.
Susceptible people. The lifetime utility of a susceptible person, Uts ,

is
Uts = u(cts ,nst )+β (1−τt )Ut+1
s i
+τt Ut+1 . (7)
Here, the variable τt represents the probability that a susceptible person
becomes infected

τt = π1 cts It Cti +π2 nst It Nti +π3 It . (8)
Critically, susceptible people understand that consuming less and working less
reduce the probability of becoming infected.
The first-order conditions for consumption and hours worked are

u1 (cts ,nst )−(1+μt )λsbt +λτ t π1 It Cti = 0,

u2 (cts ,nst )+wt λsbt +λτ t π2 It Nti = 0.
Here, λsbt and λτ t are the Lagrange multipliers associated with constraints (6)
and (8), respectively.
5157

The first-order condition for τt is

i
β Ut+1 −Ut+1
s
−λτ t = 0. (9)
Infected people. The lifetime utility of an infected person, Uti , is

Uti = u(cti ,nit )+β (1−πr −πd )Ut+1
i r
+πr Ut+1 . (10)
The expression for Uti embodies a common assumption in macro and health
economics that the cost of death is the forgone utility of life.
The first-order conditions for consumption and hours worked are given by
u1 (cti ,nit ) = λibt (1+μt ),

u2 (cti ,nit ) = −φ i wt λibt ,
where λibt is the Lagrange multiplier associated with constraint (6).
Recovered people. The lifetime utility of a recovered person, Utr , is

Utr = u(ctr ,nrt )+βUt+1
r
. (11)
The first-order conditions for consumption and hours worked are
u1 (ctr ,nrt ) = λrbt (1+μt ),
u2 (ctr ,nrt ) = −wt λrbt ,

where λrbt is the Lagrange multiplier associated with constraint (6).
Government budget constraint. The government budget constraint

is

μt St cts +It cti +Rt ctr = t (St +It +Rt ).
Equilibrium. In equilibrium, each person solves his or her maximization

problem and the government budget constraint is satisfied. The goods and
labor markets clear:
St Cts +It Cti +Rt Ctr = ANt ,
St Nts +It Nti φ i +Rt Ntr = Nt .

j j j j
In addition, ct = Ct and nt = Nt for j = s, i, and r. In the appendix, we describe
our algorithm for computing the equilibrium.7
7 Matlab replication codes can be downloaded from the authors’ websites or directly from
https://tinyurl.com/ERTcode.
5158

2. Medical Preparedness, Treatments, and Vaccination

In this section, we extend the SIR-macro model in three ways. First, we allow for
the possibility that the case fatality rate increases with the number of infections.
Second, we allow for the probabilistic development of a cure for the disease.
Third, we allow for the probabilistic development of a vaccine that inoculates
susceptible people against the virus.
2.1 The medical preparedness model

In our basic SIR-macro model, we abstract from the possibility that the efficacy
of the health care system deteriorates if a substantial fraction of the population
becomes infected. A simple way to model this scenario is to assume that the

case fatality rate depends on the number of infected people, It :
πdt = πd +κIt2 ,
where κ > 0. This functional form implies that the case fatality rate is a convex
function of the fraction of the population that becomes infected.8 The basic
SIR-macro model corresponds to the special case of κ = 0.
2.2 The treatment model

The basic SIR-macro model abstracts from the possibility that an effective
treatment against the virus will be developed. Suppose instead that an effective
treatment that cures infected people is discovered with probability δc each
period. Once discovered, treatment is provided to all infected people in the
period of discovery and in all subsequent periods, thereby transforming them
into recovered people. As a result, the number of new deaths from the disease
goes to zero.
The lifetime utility of an infected person before the treatment becomes
available is

Uti = u(cti ,nit )+(1−δc ) (1−πr −πd )βUt+1
i r
+πr βUt+1 r
+βδc Ut+1 . (12)
This expression reflects the fact that with probability δc , this person receives
treatment and becomes recovered.
We now discuss the impact of an effective treatment on population dynamics.
Before the treatment is discovered, population dynamics evolve according to
Equations (1), (2), (3), (4), and (5). Suppose that the treatment is discovered
at the beginning of time t ∗ . Then, all infected people become recovered. The
number of the deceased stabilizes once the treatment arrives, so for t ≥ t ∗ ,
Dt = Dt ∗ .
8 We do not explicitly impose the constraint that π ≤ 1, but this constraint is satisfied in all our simulations.
dt
5159

All infected people are instantly cured. Since no one dies from the disease,
we normalize the number of susceptible people to zero for t > t ∗ . The number
of recovered people is given by
Rt = 1−Dt .
2.3 The vaccination model

The basic SIR-macro model abstracts from the possibility that a vaccine against
the virus will be developed. Suppose instead that a vaccine is discovered with
probability δv per period. Once discovered, the vaccine is immediately provided
to all susceptible people.

The lifetime utility of a susceptible person is given by
Uts = u(cts ,nst )+(1−δv )(1−τt )βUt+1

s r
+δv (1−τt )βUt+1 i
+τt βUt+1 . (13)
This expression reflects the fact that with probability 1−δv , a person who is
susceptible at time t and did not get infected remains susceptible at time t +1.
With probability δv , this person is vaccinated and becomes immune to the
disease. So, at time t +1, this person’s health situation is identical to that of a
recovered person. The vaccine does not affect the health status of people who
are infected or recovered. The lifetime utilities of infected and recovered people
are given by (10) and (11), respectively.
We now discuss the impact of vaccinations on population dynamics. Before
the vaccine is discovered, these dynamics evolve according to Equations (1),
(2), (3), (4), and (5). Suppose that the vaccine is discovered at the beginning
of time t ∗ . Then, all susceptible people become recovered. Since no one is
susceptible, there are no new infections.
Denote the number of susceptible and recovered people right after a vaccine
is introduced at time t ∗ by St∗ and Rt∗ . The values of these variables are
St∗ = 0
Rt∗ = Rt ∗ +St ∗ .
For t ≥ t ∗ , we have

Rt +πr It for t = t ∗
Rt+1 =
Rt +πr It for t > t ∗ .
The laws of motion for It and Dt are given by (3) and (5).
3. Competitive Equilibrium
In this section, we discuss the properties of the competitive equilibrium via a
series of numerical exercises. In the first subsection, we describe our parameter
5160

values. In the second and third subsections, we discuss how the economy
responds to an epidemic in the SIR and SIR-macro models, respectively. In
the fourth subsection, we discuss the implications of medical preparedness. In
the fifth subsection, we discuss the effects of treatments and vaccines. Finally,
in the sixth subsection, we discuss the robustness of our results.
3.1 Parameter values

In this subsection we report our choice of parameters. We are conscious of the
considerable uncertainty about the true values of these parameters. Below, we
report the robustness of our results to using different parameter configurations.
Each time period corresponds to a week. To choose the case fatality rate,

πd , we use data from the South Korean Ministry of Health and Welfare from
March 16, 2020.9 These estimates are relatively reliable because, as of late
March, South Korea had the world’s highest per capita test rates for COVID-19
(Pueyo 2020). Estimates of case fatality rates based on data from other countries
are probably biased upward because the number of infected people is likely to
be underestimated. We compute the weighted average of the case fatality rates
using the percentage of the U.S. population in different age groups as weights.
If we exclude people that are 70 and older because their labor force participation
rate is very low, we obtain an average case fatality rate of 0.4%. If we exclude
people that are 75 and older, we obtain an average case fatality rate of 0.7%.
Based on these estimates, we set the case fatality rate equal to 0.5% and report
robustness results below. Our baseline case fatality rate is consistent with the
estimates reported in Salje et al. (2020).
As in Atkeson (2020), we assume that it takes 18 days on average to either
recover or die from the infection. Since our model is weekly, we set πr +πd =
7/18. A 0.5% case fatality rate for infected people implies πd = 7×0.005/18.
We now discuss our calibration procedure to choose the values of π1 , π2 ,
and π3 . It is common in epidemiology to assume that the relative importance of
different modes of transmission is similar across viruses that cause respiratory
diseases. Ferguson et al. (2006) argue that, in the case of influenza, 30% of
transmissions occur in the household, 33% in the general community, and 37%
in schools and workplaces.
To map these estimates into our transmission parameters, we proceed as
follows. We use the Bureau of Labor Statistics 2018 American Time Use
Survey (ATUS) to estimate the percentage of time spent on “general community
activities” that is devoted to consumption. We compute the latter as the fraction
of time spent on “purchasing goods and services” or “eating and drinking
outside the home.” To estimate the time spent “eating and drinking outside the
home,” we multiply the time spent “eating and drinking” by the fraction of
total food expenditures on “food away from home” in 2018 (54% according
9 This estimate is roughly eight times greater than the average influenza case fatality rate in the United States.
5161

to the U.S. Department of Agriculture).10 These considerations imply that the

fraction of time spent on general community activities related to consumption
activities is 48%. Since 33% of transmissions occur in the general community,
we estimate that 16% of transmissions are related to consumption (0.33×0.48,
which is roughly one-sixth).
Turning to work, recall that 37% of transmissions occur in schools and
workplaces. To compute the fraction of transmissions that occur in the
workplace, we weight the number of students by 10 and the number of workers
by 4. These weights are the average number of contacts per day at school
and work reported by Lee et al. (2010). According to the Bureau of Labor
Statistics, the number of students and workers in the population in 2018 is

76.6 million and 162.1 million, respectively. These considerations imply that
the fraction of transmissions occurring in the workplace is 46% (162.1×
4/(162.1×4+76.6×10)). Since 37% of transmissions occur in schools and
workplaces, 17% of transmissions are related to work (0.37×0.46, which is
roughly one-sixth).
We assume that virus transmission unrelated to consumption or work
activities belongs to the exogenous category (π3 St It ) emphasized in the SIR
model. The values of π1 , π2 , and π3 are chosen to satisfy
π1 C 2
= 1/6,
π1 C 2 +π2 N 2 +π3
π2 N 2
= 1/6.
π1 C 2 +π2 N 2 +π3
Here, C and N are consumption and hours worked in the preinfection steady
state. In addition, we assume that at the limit of the simple SIR model, 60%
of the population either recovers from the infection or dies. This assumption
corresponds to the Merkel scenario discussed in the introduction. The implied
values for π1 , π2 , and π3 are 7.8408×10−8 , 1.2442×10−4 , and 0.3901,
respectively.
Our calibration procedure requires various judgment calls. For example, we
had to choose which categories to include in “general community activities.”11
For this reason, we report robustness results below.
The initial population is normalized to one. The number of people who are
initially infected, ε, is 0.001. We choose A = 39.835 and θ = 0.001275 so that
in the pre-epidemic steady state, the representative person works 28 hours per
week and earns a weekly income of $58,000/52. We obtain the per capita
10 We classify the following entries in the ATUS survey as general community activities: purchasing goods and
services; eating and drinking outside the home; organizational, civic, and religious activities; socializing and
communicating; sports, exercise, and recreation; and caring for and helping nonhousehold members.
11 We chose to focus on the connection between market activities and the epidemic. This choice led us to abstract
from the response of nonmarket activities (e.g., “organizational, civic, and religious activities”) to the outbreak
of an epidemic.
5162

income estimate for 2019 from the U.S. Bureau of Economic Analysis and
the average number of hours worked from the Bureau of Labor Statistics 2018
ATUS. We set β = 0.961/52 so that the value of a life is 9.3 million 2019 dollars
in the pre-epidemic steady state. This value is consistent with the economic
value of life used by U.S. government agencies in their decision process.12 We
understand there is considerable uncertainty in the literature about this value.
We find that our conclusions are robust to reasonable perturbations of this value.
We set φ i , the parameter that controls for the relative productivity of infected
people, at 0.8. This value is consistent with the notion that symptomatic people
don’t work and the assumption that 80% of infected people are asymptomatic,
according to the China Center for Disease Control and Prevention. In the

baseline SIR-macro model, the containment rate, μt , is equal to zero.
In the medical preparedness model, we set κ equal to 0.9, which implies a
peak case fatality rate of 1%, two times higher than that in the basic SIR-macro
model. We obtain this higher case fatality rate by computing the weighted
average of the case fatality rates in Italy with weights equal to the percentage
of different age groups in the U.S. population, excluding people that are 70
and older. In both the treatment and vaccination models, we set δc = δv = 1/52,
which implies that it takes 52 weeks on average for these medical discoveries
to become available.
The model’s basic reproduction number A statistic widely used to

diagnose the severity of an epidemic is the “basic reproduction number,” R0 .
This statistic is the total number of infections caused by one infected person
(with measure zero) in his or her lifetime in a population in which everybody
is susceptible (S0 = 1). The higher is the value of R0 , the faster is the spread of
the virus.
The average rate of infection, which we denote by γ , is the ratio of the
number of newly infected people to the total number of infected people at the
beginning of an epidemic (T0 /I0 ). The expected number of infections caused
by a single infected person is
γ
R0 = γ +(1−πr −πd )γ +(1−πr −πd )2 γ +... = .
πr +πd
In this expression, (1−πr −πd )t is the probability that the infected person
reaches period t without recovering or dying.
In the epidemiology literature, the value of R0 is generally estimated using
one of two methods (see, e.g., Breban, Vardavas, and Blower 2007). The first
method uses individual-level data collected by contact tracing at the beginning
of the epidemic to estimate the number of secondary infections produced by
an infected person. The second method involves choosing a value of R0 so
12 See U.S. Environmental Protection Agency (2010) and Moran (2016). See Viscusi and Aldy (2003) for a review
of the literature on the value of a statistical life.
5163

that a given model matches aggregate data on the number of infections and
deaths during an epidemic episode. The implied estimates of R0 depend on the
features of the model, including parametric assumptions. So, reported standard
errors in any given study understate the true uncertainty about R0 .13
We proceed in the spirit of the second method and choose parameters that
imply a value of R0 = 1.45 because they produce plausible implications for the
dynamics of the epidemic. This value is at the low end of available estimates for
R0 but consistent with the evidence taking sampling uncertainty into account
(see, e.g., Riou and Althaus 2020). In Subsection 4.6, we discuss the robustness
of our results to different values of R0 .

3.2 The SIR model
The dashed pink lines in Figure 1 represent the equilibrium population dynamics
implied by the SIR model. The share of the initial population who is infected
peaks at 6.8% in week 31. Thereafter, this share falls because fewer people are
susceptible to infection. Eventually, 60% of the population becomes infected.
Assuming a U.S. population of 330 million people, this scenario implies that
roughly 200 million Americans eventually become infected. A case fatality
rate of 0.5% implies that the virus kills roughly 1 million people in the United
States.
Figure 1 shows that the epidemic induces a recession: aggregate consumption
falls by roughly 1.5% from peak to trough. This fall reflects two factors. First
and foremost, the virus causes infected people to be less productive at work (φ i =
0.8). The associated negative income effect lowers the consumption of those
who are infected. The dynamic behavior of aggregate consumption mimics the
share of infected people in the overall population. Second, the death toll caused
by the epidemic permanently reduces the size of the workforce.
Since the production function has constant returns to scale, per capita income
is the same in the pre- and post-epidemic steady states. In the post-epidemic
steady state, the population and real gross domestic product (GDP) are both
0.3% lower than in the initial steady state.
3.3 The SIR-macro model

In the SIR model, economic decisions about consumption and work don’t
influence the dynamics of the epidemic. In contrast, in the SIR-macro model,
susceptible people can lower the probability of being infected by reducing their
consumption and hours worked. The solid blue lines in Figure 1 depict how the
epidemic unfolds in this model.
The share of the initial population who is infected peaks at 5.3% in week 33.
This peak is substantially smaller and occurs a little later than the corresponding
13 The difficulty in estimating R is reflected in the broad range of estimates obtained for widely studied diseases,
0
such as measles. The literature on the measles epidemics often cites values of R0 from 12 to 18. In a recent
survey, Guerra et al. (2017) find an even wider range of R0 estimates, from 4 to 60.
5164

Basic SIR-Macro Model SIR Model ( 1 = 2 =0, model recalibrated)
Infected, I Susceptibles, S Recovered, R

8 100 60
% of Initial Population
90 50
6
80 40
4 70 30
60 20
2
50 10
0 40 0
0 50 100 150 0 50 100 150 0 50 100 150

Deaths, D Aggregate Consumption, C Aggregate Hours, N
% Dev. from Initial Steady State

0.3 0 0
0.25 -2 -2
0.2
-4 -4
0.15
-6 -6
0.1
0.05 -8 -8
0 -10 -10
0 50 100 150 0 50 100 150 0 50 100 150
Weeks Weeks Weeks
Figure 1
Basic SIR-Macro Model vs. SIR Model
peak in the SIR model. Eventually, 54% of the population becomes infected. So,
for the United States, roughly 180 million people eventually become infected
and 890,000 people die.
Figure 1 shows that the infection is less severe in the SIR-macro model than
in the SIR model. The reason is that in the SIR-macro model, susceptible people
severely reduce their consumption and hours worked to lower the probability of
being infected. Figure 2 shows that no offsetting effects arise from the behavior
of recovered and infected people because they behave as in the SIR model.
Consistent with these observations, the recession is much more severe in
the SIR-macro model: average aggregate consumption in the first year of the
epidemic falls by 4.7%, a fall seven times larger than in the SIR model.
The dynamics of hours worked are similar in the two models but the
magnitudes are very different.
In the SIR-macro model, hours worked follow a U-shaped pattern. The peak
decline of 9.8% occurs in week 33. Thereafter, aggregate hours rise, converging
to a new steady state from below. These dynamics are driven by the labor
supply decisions of susceptible people. Interestingly, the long-run decline in
hours worked is slightly lower in the SIR-macro model (0.27%) than in the
5165

Consumption by Type Hours by Type
0 0
-5 -5

-10 -10

-15 -15
Consumption Susceptibles
Consumption Infected
Consumption Recovered
Hours Susceptibles
-20 -20 Hours Infected
Hours Recovered
0 50 100 0 50 100
Weeks Weeks
Figure 2
Consumption and Hours by Type in Basic SIR-Macro Model
SIR model (0.30%). The reason is that fewer people die in the epidemic, so the
population falls by less in the SIR-macro model than in the SIR model.
Figure 3 shows the competitive equilibrium and the optimal containment
policy in the SIR-macro model. We return to this figure in the next section.
3.4 Medical preparedness model

The dashed dotted red lines in Figure 4 show that the competitive equilibrium
with an endogenous case fatality rate involves a much larger recession than
that in the basic SIR-macro model (solid blue lines). The reason is that people
internalize the higher case fatality rates associated with a health care system that
can become overburdened with infected people. Since the costs of becoming
infected are much higher, people cut back more on consumption and work to
reduce the probability of becoming infected. The net result is that fewer people
are infected but more people die.
3.5 The treatment and vaccination models

As discussed in the introduction, the possibility of treatments being discovered
makes people become more willing to engage in market activities. The reason
is that the expected costs of being infected are smaller. Because of this change
in behavior, the recession is less severe. In Figure 5 the solid blue and dashed
5166

Basic SIR-Macro Model Best Simple Containment Policy Simple Command Containment Policy

6 100 60
5 90 50
4 80 40
3 70 30
2 60 20
1 50 10
0 40 0
0 50 100 150 0 50 100 150 0 50 100 150
Deaths, D Aggregate Consumption, C Best Containment Policy,

0.3 0 80
0.25 -5
60
0.2 -10
%
0.15 -15 40
0.1 -20
20
0.05 -25
0 -30 0
0 50 100 150 0 50 100 150 0 50 100 150
Weeks Weeks Weeks
Figure 3
Basic SIR-Macro Model with and without Containment
dotted red lines virtually coincide. So, in practice, the quantitative effect of
the possibility of treatments on the competitive equilibrium is quite small. As
discussed in Section 5.4, the possibility of treatments does not substantively
affect the design of optimal policy.
Vaccines don’t cure infected people, but they do prevent susceptible people
from becoming infected. So the possibility of future vaccination induces people
to reduce their market activities before the vaccine actually arrives. Given our
calibration, this effect is small in the competitive equilibrium (see the solid blue
and dashed dotted red lines in Figure 6). But, as discussed in Section 5.4, the
possibility of vaccinations substantively affects the design of optimal policy.
It becomes optimal to immediately introduce severe containment measures to
minimize deaths.
3.6 Robustness
Table 1 reports results of a series of robustness exercises in which we vary
key parameters of the basic SIR-macro model. Consider first the parameter
φ i , which controls the productivity of infected workers. The lower is φ i ,
the smaller is the average consumption drop, the peak infection rate, the
cumulative mortality rate, and the total number of U.S. deaths. The behavior
of aggregate consumption reflects two opposing forces. On the one hand, a
lower φ i makes it more costly to become infected. So, susceptible people
5167

Basic SIR-Macro Model ( d

constant) Endog. Case Fatality Rate ( d
= f ( Infected ) ) Best Simple Containment Policy

% of Ini. Pop. 6 100 60
% of Ini. Pop.
% of Ini. Pop.
4 80 40
2 60 20
0 40 0
0 50 100 150 0 50 100 150 0 50 100 150
Deaths, D Aggregate Consumption, C Aggregate Hours, N

0.4
0 0
% of Ini. St.St.
% of Ini. St.St.
% of Ini. Pop.
-10 -10
0.2
-20 -20
-30 -30

0
0 50 100 150 0 50 100 150 0 50 100 150
Case Fatality Rate, Weeks
d Best Containment Policy,
1 100
0.8
%
50
0.6
0.4 0
0 50 100 150 0 50 100 150
Weeks Weeks
Figure 4
Medical Preparedness
reduce their consumption by more. On the other hand, cautious behavior by

susceptible people reduces the total number of people infected. Since infected
people consume much less than susceptible people (see Figure 2), this effect
increases average consumption in the population. In our model, the first force
is somewhat stronger than the second.
Table 1 also reports the results for different parameters of the infection
transmission function (Equation (1)). Recall that in the benchmark model,
we choose our baseline parameters so that, at the beginning of the infection
episode, economic decisions account for one-third of the infection rate. Table 1
summarizes results for the case in which economic decisions account for one-
sixth of the initial infection rate. In this scenario, the drop in consumption is
smaller. The reason is that people understand that economic activity has less of
an impact on infection rates. The peak infection rate, the cumulative mortality
rate, and the total number of U.S. deaths are larger. Table 1 also reports the case
in which economic decisions account for two-thirds of the initial infection rate.
In this scenario, the drop in consumption is larger, and the peak infection rate
and cumulative mortality rate are smaller. The reason is that people cut back
more on economic activities because they have a larger impact on infection
rates.
Next, we increase the case fatality rate from 0.5% to 1%. This change
increases the severity of the recession as people cut back on their consumption
5168

Basic SIR-Macro Model Model with Treatment Best Simple Containment Policy with Treatment

6 100 60
5 90 50
4 80 40
3 70 30
2 60 20
1 50 10
0 40 0
0 50 100 150 0 50 100 150 0 50 100 150
Deaths, D Aggregate Consumption, C

Best Containment Policy,
0.3 0 80
0.25 -5
60
0.2 -10
%
0.15 -15 40
0.1 -20
20
0.05 -25
0 -30 0
0 50 100 150 0 50 100 150 0 50 100 150
Weeks Weeks Weeks
Figure 5
SIR-Macro Model with Treatments
and work to reduce the chances of being infected. Despite the concomitant fall
in peak infection rates, the cumulative mortality rate and the number of U.S.
deaths rise.
Table 1 reports the impact of a change in the medical preparedness parameter,
κ. The lower is κ, the higher is the degree of medical preparedness. We consider
a value of κ = 0.9 such that the case fatality rate in the medical preparedness
model peaks at 1%. Table 1 shows that this higher value of κ is associated with
a more severe recession as people curtail their economic activity in response to
higher case fatality rates. While the peak level of infections falls, the cumulative
death rate and the total number of U.S. deaths rise.
We also assess the impact of reducing the discount factor from 0.961/52 to
0.941/52 . This parameter change reduces the value of a life from 9.3 million to
6.1 million 2019 dollars. As a result, consumption falls less during the epidemic
and infection rates rise. The overall quantitative sensitivity is small.
Overall, Table 1 indicates that the qualitative conclusions of the basic SIR-
macro model are very robust and that the quantitative conclusions are robust to
the perturbations that we consider.
We now discuss the impact of different values of R0 on the properties of our
model. Recall that in the baseline SIR-macro model, R0 is equal to 1.45. Table 2
reports results for alternative values of R0 , ranging from 1.33 to 2.95. Three key
5169

Basic SIR-Macro Model Model with Vaccines Best Simple Containment Policy with Vaccines

% of Initial Population 6 100 60
5 90 50
4 80 40
3 70 30
2 60 20
1 50 10
0 40 0
0 50 100 150 0 50 100 150 0 50 100 150

0.3 0 50
0.25 40
-5
0.2 30
%
0.15 -10 20
0.1 10
-15
0.05 0
0 -20 -10
0 50 100 150 0 50 100 150 0 50 100 150
Weeks Weeks Weeks
Figure 6
SIR-Macro Model with Vaccines
features emerge from this table. First, high values of R0 generate implausibly
large peak infection rates and mortality rates. Second, high values of R0 imply
that the epidemic runs its course very quickly. For example, for R0 = 2.95,
infections peak in the 13th week of the epidemic. Third, the peak-to-trough
drop in consumption is increasing in R0 . This result reflects people’s response
to the higher probability of becoming infected. Interestingly, the average drop
in consumption over the first year of the epidemic is not very sensitive to R0 .
This property results from two effects. The first effect is a much larger peak-
to-trough drop in consumption for high values of R0 . The second effect is a
shorter epidemic, and a shorter recession, for high values of R0 . For example,
for R0 = 2.95, consumption initially drops precipitously but recovers after 13
weeks as infections wane.
In sum, Table 2 shows that the qualitative features of our model are very
robust to different values of R0 . But the quantitative properties of the model
do depend on R0 . As we discuss above, much of the evidence on R0 in the
literature is model based. Viewed through the lens of our model, the most
plausible value of R0 is relatively low, around 1.5, a value that is consistent
with the literature taking sampling uncertainty into account. If one insists on
calibrating the model with a high value of R0 , then the model must be extended
to make it consistent with the data.
5170

Table 1
Robustness in basic SIR-macro model without containmenta
Consumption Infection rate Mortality rate U.S. deaths
%b %c %d millionse
Productivity of infected people, φ i
0.7 −4.61 4.85 0.26 0.85

0.8 (baseline) −4.66 5.23 0.27 0.88
Share of initial infections due to consumption, work, and general contacts
1/12,1/12,5/6 −2.77 6.15 0.287 0.94

1/6,1/6,2/3 (baseline) −4.66 5.23 0.267 0.88
1/3,1/3,1/3 −7.24 3.25 0.218 0.72

Mortality rate, πd
0.005×7/18 (baseline) −4.66 5.23 0.26 0.88

0.01×7/18 −8.25 4.74 0.51 1.69
Limited health care capacity parameter, κ (slope of endogenous mortality rate)
0 (baseline) −4.66 5.23 0.26 0.88

0.9 −6.83 4.71 0.39 1.31
Household discount factor, β
0.961/52 (baseline) −4.66 5.23 0.26 0.88

0.941/52 −3.37 5.42 0.27 0.89
a See Section 3.6 for a discussion of the results provided in this table.
b Average drop in consumption in first year relative to preinfection steady state.
c Peak infection rate relative to pre-epidemic population.
d Cumulative mortality rate at the end of the epidemic.
e Total number of deaths in the United States at the end of the epidemic.
Table 2
Robustness in basic SIR-macro model without containment with respect to R0 a
Pop. Cons. Inf. Mort. U.S. Time to
infected Cons. trough rate rate deaths peak infection
%b R0 %c %d %e %f millionsg weekh
50 1.33 −3.4 −6.7 3.2 0.21 0.72 42

60 (baseline) 1.45 −4.7 −9.8 5.2 0.26 0.88 34
70 1.62 −5.2 −13.3 8.2 0.31 1.05 28
80 1.86 −5.2 −16.9 12.5 0.37 1.22 23
93 2.49 −4.1 −20.6 23.9 0.44 1.46 16
97 2.95 −3.2 −20.9 31.2 0.47 1.55 13
a See Section 3.6 for a discussion of the results provided in this table.
b Percentage of the population eventually infected in the canonical SIR model.
c Average drop in consumption in the first year relative to the preinfection steady state.
d Peak-to-trough drop in weekly consumption.
e Peak infection rate relative to pre-epidemic population.
f Cumulative mortality rate at the end of the epidemic.
g Total number of deaths in the United States at the end of the epidemic.
h Week at which the fraction of the infected population as a percentage of initial population peaks.
5171

The model can be extended in at least three ways. The first is to

explicitly model the impact of nonpharmaceutical interventions, such as
masks and social distancing which slow down the transmission of the virus
(see, e.g. Eichenbaum, Rebelo, and Trabandt 2020a). The second is to
include the possibility of substitution from high- to low-contact forms of
consumption and work, so that a given reduction in the transmission rate
results in a smaller decline in economic activity than in our benchmark model
(see Jones, Philippon, and Venkateswaran 2021 and Krueger, Uhlig, and Xie
2020 for models along these lines). The third is to endogenize the time that
people spend on nonmarket social interactions. In this setting, a decline in these
interactions can reduce the rate of virus transmission without producing a drop

in measured output (see Farboodi, Jarosch, and Shimer 2020 for a model along
these lines). Evaluating the empirical performance of these different approaches
is an interesting topic for future research.
4. Economic Policy
The competitive equilibrium of our model economy is not Pareto optimal. A
classic externality is associated with the behavior of infected people. Because
each person is atomistic, people don’t take into account the impact of their
actions on the infection and death rates of other people. But collectively, the
behavior of infected people imposes an unpriced cost on susceptible people.
In this section, we consider a simple Ramsey problem designed to deal with
this externality. As it turns out, the solution to the Ramsey problem is quite
similar to the solution of a planner’s problem in which the planner chooses
consumption and labor subject to the constraint that these choices are the same
for everybody regardless of health status.
4.1 Ramsey problem

As with any Ramsey problem, we must take a stand on the policy instruments
available. In reality, governments can reduce social interactions in many ways.
Examples of containment measures include shelter-in-place laws and shutting
down restaurants and bars. Analogous to Farhi and Werning’s (2012) treatment
of capital controls, we model these measures as a tax on consumption, the
proceeds of which are rebated lump sum to people in the economy. We refer to
this tax as the “containment rate.”
We compute the optimal sequence of 250 containment rates {μt }249 t=0 that
maximize social welfare, U0 , defined as a weighted average of the lifetime
utility of different people. Since at time zero R0 = D0 = 0, the value of U0 is
U0 = S0 U0s +I0 U0i . (14)

Given the sequence of containment rates, we solve for the competitive
equilibrium and evaluate the social welfare function. We iterate on this sequence
until we find the optimum.
5172

Figure 3 displays our results. First, the gradual escalation of containment

measures over time is optimal. The optimal containment rate rises from 4.5%
in week 0 to a peak value of 72% in week 37. The rise in containment rates
roughly parallels the dynamics of the infection rate itself. The basic intuition
is as follows. Containment measures internalize the externality caused by
the behavior of infected people. So, as the number of infected people rises,
intensifying the containment measures is optimal. For example, at time zero,
very few people are infected, so the externality is relatively unimportant. A
high containment rate at time zero would have a high social cost relative to the
benefit. As the infection rate rises, the externality becomes important, and the
optimal containment rate rises.

The optimal containment policy greatly reduces the peak level of infections
from 5.3% to 3.2%, thereby reducing the death toll from 0.27% to 0.21%
of the initial population. For a country like the United States, this reduction
represents roughly 200,000 lives saved. This beneficial outcome is associated
with a much more severe recession. The fall in average aggregate consumption
in the first year of the epidemic more than triples, going from about 4.7%
without containment measures to about 17% with containment measures. The
mechanism underlying this result is straightforward: higher containment rates
make consumption more costly, so people cut back on the amount they consume
and work.
Why not choose initial containment rates that are sufficiently high to induce
an immediate, persistent decline in the number of infected people? Absent
vaccines, the only way to prevent a recurrence of the epidemic is for enough
of the population to acquire immunity by becoming infected and recovering.
The optimal way to reach this critical level of immunity is to gradually increase
containment measures as infections rise and slowly relax them as new infections
wane.
4.2 Simple command containment

One possible objection to our simple containment policy is that it is modeled as a
Pigouvian consumption tax. An alternative formulation is to consider a planning
problem in which the government chooses consumption and hours worked
to maximize the objective function (14) subject to the population dynamics
equations, the resource constraint, and the constraint that people have the same
allocation regardless of health status:
Cts = Cti = Ctr ,

and
Nts = Nti = Ntr .
The solution to this “simple command containment” problem is represented
by the dash-dotted red line in Figure 3. This figure shows that the solution to
this problem is very similar to the simple containment policy discussed above.
5173

A similar conclusion holds for optimal policy in the benchmark SIR model
discussed below.14
4.3 Medical preparedness model

Comparing Figures 3 and 4, we see that the optimal containment policy is more
aggressive in the medical preparedness model than in the basic SIR-macro
model. The peak containment rate is higher in the medical preparedness model
(110% vs. 72%) and occurs earlier (at week 33 vs. week 37). In addition, the
containment rate comes down much more slowly in the medical preparedness
model. These differences reflect that, other things equal, the social cost of the
externality is much larger. People do not internalize the cost of consumption

and work on infection rates nor do they internalize the aggregate increase in
case fatality rates.
The optimal containment policy greatly reduces the peak level of infections
from 4.7% without containment to 2.2% with containment. The death toll falls
from 0.40% to 0.22% of the initial population. For a country like the United
States, this reduction represents roughly 600,000 lives saved.
4.4 The treatment and vaccination models

Comparing Figures 3 and 5, we see that the optimal containment policies in the
treatment and basic SIR-macro models are quite similar. In the treatment model,
along a path where no treatment is discovered, the optimal containment policy
reduces the peak level of infections from 5.3% to 3.2%, reducing the death toll
from 0.27% to 0.21% of the initial population. This reduction corresponds to
roughly 200,000 lives saved in the United States. The latter figure pertains to
a worst-case scenario in which a treatment is never discovered.
The dashed black lines in Figure 6 show that optimal policy is very different
in the basic SIR-macro model and the vaccination model. With vaccines as a
possibility, it is optimal to immediately introduce severe containment measures
to minimize the number of deaths. Those containment measures cause a very
large, persistent recession: average consumption in the first year of the epidemic
falls by about 17%. But this recession is worth incurring in the hope that the
vaccines arrive before many people get infected.
It is optimal to reduce and delay the peak of the infections in anticipation of
a vaccine being discovered. Figure 6 displays the behavior of the vaccination
model under optimal containment policy on a path in which a vaccine does not
14 Both the peak in infections and the fall in aggregate consumption are slightly larger under simple command
containment than under simple containment. This result reflects compositional effects. First, the consumption
of recovered people drops by more under simple command containment because everybody must have the
same consumption. Second, the consumption of infected people is substantially lower under simple containment
than under simple command containment. This property reflects infected people’s lower productivity and their
response to the Pigouvian tax. Third, the consumption of susceptible people is similar under both policies. The net
effect is that, even though aggregate consumption drops by more under simple command containment, infection
rates are higher, reflecting the higher consumption of infected people.
5174

arrive. Compared to the competitive equilibrium (dashed dotted red lines), the
peak of the infection rate drops from 5.3% to 3.3% of the initial population.
Moreover, the infection peak occurs in week 42 rather than in week 33. Absent
a vaccine being discovered, the optimal containment policy reduces the death
toll as a percentage of the initial population from 0.27% to 0.24%. For the
United States, this reduction amounts to about 100,000 lives. It is important to
remember that this reduction pertains to a worst-case scenario in which vaccines
do not arrive.
Above, we discuss why it is not optimal to introduce immediate containment
measures in the basic SIR-macro and treatment models. But why is the optimal
policy so different in the vaccination model? The basic reason is that, unlike

curative treatments, a vaccine does not cure infected people. In addition, the
expected arrival of a vaccine also reduces the importance of achieving herd
immunity before the vaccine arrives.
5. Quantitative Predictions for the Benchmark Model

In the previous sections, we separately analyze the quantitative predictions of
our model under different simplifying assumptions. Those exercises are useful
for understanding the mechanisms at work. In our view, the most meaningful
version of the model allows for both (a) the possibility of vaccines and medical
treatment and (b) the impact of the number of infections on the efficacy of the
health care system. We refer to this version of the model as the benchmark
model. We describe the equations used to compute the competitive equilibrium
of this model in the Appendix.
In the first subsection, we discuss the impact of optimal containment policy
in the benchmark model. The second subsection considers the consequences
of prematurely ending the containment policy. In the third subsection, we
consider the implication of a delay in implementing the optimal containment
policy. In the fourth subsection, we study a first-best solution in which the
planner chooses directly different levels of consumption and hours worked for
susceptible, infected, and recovered people.
5.1 Optimal policy in the benchmark model

The solid blue and dashed black lines in Figure 7 represent the evolution of the
economy in the competitive equilibrium and under the best simple containment
policy, respectively. Consistent with previous figures, we display a path along
which vaccines and treatments are not discovered.
From a qualitative point of view, the benchmark model inherits key features
of its underlying components. Consistent with the vaccination model, it is
optimal to immediately introduce severe containment (μ = 43%). Consistent
with the treatment and medical preparedness models, it is optimal to ramp up
containment as the number of infections rises. The maximal containment rate
reaches 76% in week 32.
5175

The optimal containment measures substantially increase the severity of the

recession. Without containment, average consumption in the first year of the
epidemic falls by about 7%. With containment, this fall is 22%. Notably, the
size of the recession is smaller than in the medical preparedness model. The
reason is that the prospect of vaccinations and treatments reduces the magnitude
of the externality associated with the medical preparedness problem.
The benefit of the large recession associated with optimal containment in
the combined model is a less severe epidemic. Compared to the competitive
equilibrium, the peak infection rate drops from 4.7% to 2.5% of the initial
population. The optimal policy reduces the death toll as a percentage of the
initial population from 0.40% to 0.26%. For the United States, this reduction

amounts to about half a million lives.
We emphasize that the latter reduction pertains to a worst-case scenario in
which vaccines and treatments never arrive. If they do arrive, many more lives
would be saved. Thankfully, they would be saved by medicine rather than by
containment policies.
The dashed dotted red lines in Figure 7 represent the optimal containment
policy implemented as a solution to a planner’s problem in which the planner
chooses consumption and hours worked subject to the constraint that people
have the same allocation regardless of health status. We see that the solution
under this “command containment” policy is very similar to the solution under
the best simple containment policy.
5.2 The costs of ending containment too early

As a practical matter, policy makers could face intense pressure to prematurely
end containment measures because of their negative impact on economic
activity. In this subsection, we discuss the costs of doing so. The solid red
lines in panels A and B of Figure 8 represent the response of the economy
to an unanticipated end of the best simple containment policy after weeks
12 and 44, respectively. Week 44 is when infections peak under the best
simple containment policy. The dashed black lines represent the behavior of
the economy when the best simple containment policy is fully implemented.
From panel A, we see that abandoning containment initially generates a large
recovery, with consumption surging by roughly 17%. Unfortunately, this surge
results in a large rise in infection rates. The latter rise plunges the economy into
a second, persistent recession.
So, prematurely abandoning containment produces a temporary rise in
consumption but no long-lasting economic benefits. Tragically, abandonment
leads to a substantial rise in the death toll of the epidemic.
Panel B shows that the longer policy makers pursue the optimal containment
policy, the better. Both the temporary gains and the losses due to abandoning
the optimal policy in panel B are smaller than those in panel A.
The implications of our model for the cost of ending containment too early
are consistent with the evidence for the 1918 Spanish flu (Bootsma, Martin, and
5176

Benchmark SIR-Macro Model Best Simple Containment Policy Simple Command Containment Policy

5 100 60
90 50
4
80 40
3
70 30
2
60 20
1 50 10
0 40 0
0 50 100 150 0 50 100 150 0 50 100 150

0.4 0 80
-5
60
0.3
-10
40
%
0.2 -15
20
-20
0.1
-25 0
0 -30 -20
0 50 100 150 0 50 100 150 0 50 100 150
Weeks Weeks Weeks
Figure 7
Benchmark SIR-Macro Model (Vaccines, Treatment, Med. Preparedness)
Ferguson 2007). We conclude that it is important for policy makers to resist

the temptation to pursue transient economic gains obtained by abandoning
containment measures.
5.3 Costs of starting containment too late

Policy makers can also face pressure to delay implementing optimal
containment measures. The dashed dotted red lines in Figure 9 represent the
impact of beginning containment only in week 33, the week in which infections
peak. We assume that the best simple containment policy is calculated and
implemented from that point on. The dashed black lines represent the behavior
of the economy when the best simple containment policy is implemented from
week 0 on. The solid blue line represent the competitive equilibrium with no
containment measures.
The best simple containment policy that begins in week 33 involves draconian
containment measures that cause an enormous drop in economic activity.
The reason is simple: with infections raging, the externalities associated with
economic activity are very large.
Despite the draconian measures, the total number of deaths caused by
the epidemic is much larger than if the best simple containment policy is
implemented without delay. Still, as far as the death toll of the epidemic
5177

Best Simple Containment Policy in Benchmark Model Early Exit from Best Simple Containment Policy
Infected, I Deaths, D Agg. Consumption, C Best Containment Policy,

5 0.4 80
Panel A: Exit after 12 Weeks
4 -5 60
0.3
3 -10
40
0.2 -15
2
-20 20
0.1
1
-25
0
0 0 -30
0 50 100 0 50 100 0 50 100 0 50 100
Infected, I Deaths, D Agg. Consumption, C Best Containment Policy,

5 0.4 80
Panel B: Exit after 44 Weeks
4 -5 60
0.3
3 -10
40
0.2 -15
2
-20 20
0.1
1
-25
0
0 0 -30
0 50 100 0 50 100 0 50 100 0 50 100
Figure 8
is concerned, late containment (dashed dotted red lines) is better than no

containment at all (solid blue lines).
The implications of our model for the cost of starting containment too late
are consistent with the evidence for the 1918 Spanish flu (Hatchett, Carter,
and Lipsitch 2007; Bootsma, Martin, and Ferguson 2007). We conclude that
it is important for policy makers to resist the temptation to delay containment
measures for the sake of initially higher short-run levels of economic activity.
5.4 Smart containment

Recall that in Section 4.1 we study simple containment policies corresponding
to a Ramsey problem in which the government chooses the same consumption
containment rate for everybody in the economy. In Section 4.2 we study the
optimal simple command containment policy in which the government chooses
the same consumption and hours worked for all health types.
In this section, we study smart containment, by which we mean the solution
to a social planning problem in which the planner directly chooses different
levels of consumption and hours worked for susceptible, infected, and recovered
people.
The planner maximizes the social welfare, U0 , defined in equation (14). The
planner chooses Cts , Cti , Ctr , Nts , Nti , and Ntr for all t to maximize U0 subject to
5178

Benchmark SIR-Macro Model Best Simple Containment Policy Late Start of Best Simple Containment Policy (Week 33)

5 100 60
90 50
4
80 40
3
70 30
2
60 20
1 50 10
0 40 0
0 50 100 150 0 50 100 150 0 50 100 150

0.4 0 250

-10 200
0.3
-20 150
%
0.2 -30 100
-40 50
0.1
-50 0
0 -60 -50
0 50 100 150 0 50 100 150 0 50 100 150
Weeks Weeks Weeks
Figure 9
the expressions for the lifetime utility of the different people, the transmission
function (1), and the laws of motion for the population, (2), (3), (4), and (5).
The lifetime utilities of susceptible, infected and recovered people are
given by
Uts = u(Cts ,Nts )+(1−δv )(1−Tt )βUt+1

s r
+δv (1−Tt )βUt+1 i
+Tt βUt+1 ,

Uti = u(Cti ,Nti )+(1−δc ) (1−πr −πdt )βUt+1
i r
+πr βUt+1 r
+βδc Ut+1 ,
Utr = u(Ctr ,Ntr )+βUt+1

r
.
The lifetime utility of susceptible people is computed using the aggregate
transition probabilities because the planner internalizes the infection
externality.
Figure 10 summarizes our results. Note that infected people do not work
unless they recover. As a result, all susceptible people can work without fear
of becoming infected. The planner sets the consumption of infected people to
a minimum. In fact, there is no maximum to the social planning problem, only
a supremum. Given the form of momentary utility, it is not optimal to set the
consumption of infected people to zero because their utility would be equal to
5179

Infected, I Susceptibles, S
0.12 100
0.1
99
0.08
98
0.06
97
0.04
96
0.02
0 95
0 100 200 300 400 500 0 100 200 300 400 500
Deaths, D Agg. Consumption, C

0.025 0

-0.02
0.02
-0.04
0.015
-0.06
0.01
-0.08
0.005
-0.1
0 -0.12
0 100 200 300 400 500 0 100 200 300 400 500
Weeks Weeks
Figure 10
Smart Containment in the Benchmark SIR-Macro Model
−∞. But the closer the consumption of those infected is to zero, the higher
is social welfare. Because infected people are completely isolated, the initial
infection quickly dies out without causing a recession.
The previous analysis assumes that infected people have to be in contact
with other people to obtain consumption goods. This assumption underlies the
draconian policy implication that the consumption of infected people should
be kept at a minimum. Suppose instead that the planner can directly deliver
consumption goods to the infected so they do not need to go shopping. The
solution to this modified problem continues to have the property that infected
people don’t work. But they consume the same as other people. Since there is
such a small number of infected people at time zero, aggregate consumption
and hours worked are essentially the same as in the pre-epidemic steady state.
One important limitation of smart containment is that herd immunity is never
reached. Every infection case must be detected and dealt with until the virus
is extinguished. Otherwise, the epidemic would begin anew as soon as smart
containment ended.
Our simple analysis of smart containment assumes that policy makers
know the health status of different individuals. In reality, this knowledge
would require antigen and antibody tests for immunity and infection that are
sufficiently accurate to act on. Our results suggest enormous social returns to
5180

having these tests and the policy instruments to implement smart containment.15
This conclusion is consistent with the importance of early detection and early
response emphasized by epidemiologists, such as Ginsberg et al. (2008). In
a subsequent paper, Eichenbaum, Rebelo, and Trabandt (2020a) study the
efficacy of smart containment policies in a setting in which people do not
know their health status until they are tested.
6. Model Performance in Retrospect

This paper was originally drafted in March 2020. Almost a year has passed,
and a natural question now is: how do the predictions of the model compare to
currently available data?

The first-order prediction of the model is that, in response to the epidemic,
economic activity would undergo a sharp contraction followed by a robust
recovery. This qualitative prediction is clearly supported by data for the U.S. and
other developed economies. The model also does reasonably well at capturing
the broad quantitative decline in economic activity. For example, the peak-
to-trough decline in U.S. real GDP (between the last quarter of 2019 and the
second quarter of 2020) is 10.7%. In our model, the analogue decline in output,
absent containment, is 5.4%. The average decline in U.S. real GDP in 2020
relative to 2019 is 3.6%. In the model, the analogue decline in output, absent
containment, is 4.7%. Going beyond these broad statements and providing a
detailed comparison of model and data would require taking a stand on the
timing and nature of the containment measures implemented in the United
States.
Another important implication of our model is that economic activity
should contract even before the government imposes containment measures.
Substantial evidence supports this prediction. See, for example, the results in
Goolsbee and Syverson (2021) and Villas-Boas et al. (2020), based on mobility
data, as well as the evidence in Chetty et al. (2020) and Eichenbaum et al. (2020),
based on micro data on consumption.
The implications of our model are also consistent with evidence presented
by Atkeson, Kopecky, and Zha (2020) for a large number of countries and U.S.
states. These authors show that the growth rate of daily COVID-19 deaths fell
much more rapidly than predicted by a canonical SIR model, suggesting strong
behavioral changes that slowed down the transmission of the virus.
Our model predicts that, absent a successful vaccination campaign, the
death toll of the epidemic would be between 500,000 and 1.5 million people,
depending on the containment policy adopted. As of February 6, 2021, 460,000
Americans have died of COVID-19. On February 4, 2021, the University of
Washington’s Institute for Health Metrics and Evaluation predicted that the
15 According to de Walque, Friedman, and Mattoo (2020), the average cost per test, including equipment,
consumables, protective equipment, and labor, ranges from $2 to $5.
5181

total death toll will reach 630,000 by June 2021. This forecast lies within the
range of the model’s predictions although somewhat at the lower end, reflecting
in part the expected positive impact of the ongoing vaccination campaign.
An important failing of the model has to do with the timing of these deaths.
In the data, a significant second wave of deaths took place between December
2020 and February 2021. Our model generates a second wave if we assume
that containment ended prematurely (see Figure 8). But the magnitude of that
wave is much smaller than that observed in the data. We do not think that this
failing reflects the calibration that we used. Consistent with Atkeson (2021),
we conjecture that a combination of seasonality in the transmission rate and
changes in the way people react to infection risk are likely to be important

factors in accounting for the second wave of COVID-19 deaths.
Finally, our model predicts that, absent instant mass vaccination, the
epidemic episode would last between 18 months and two years. This prediction
seems reasonable in light of the fact that the episode is not yet over but deaths
are now declining. But, of course, now that vaccines are available, the end of
the epidemic depends on the pace of the vaccination campaign and the efficacy
of vaccines in the face of ongoing mutations of the COVID-19 virus.
7. Related Literature
Our work is related to a pre-Covid literature that combines economics and
epidemiology (for a review, see Perrings et al. 2014). Examples include analyses
of how private vaccination incentives affect epidemic dynamics and optimal
public health policy (e.g., Philipson 2000; Manski 2010; Adda 2016) and studies
of the interaction between behavioral choice and the dynamics of the HIV/AIDS
epidemic (e.g., Kremer 1996; Greenwood et al. 2019).
The COVID-19 crisis has stimulated a rapidly growing body of work on the
economics of the epidemic. Below, we briefly summarize the first wave of this
research program.
Atkeson (2020) provides an overview of SIR models and explores their
implications for the COVID-19 epidemic. Alvarez, Argente, and Lippi (2020)
study the optimal lockdown policy in a version of the canonical SIR model
in which the case fatality rate increases with the number of infected people.
Toxvaerd (2020) analyses the equilibrium amount of social distancing in a SIR
model and argues that it is not socially optimal.
Jones, Philippon, and Venkateswaran (2021) study optimal mitigation
policies in a model in which economic activity and epidemic dynamics interact.
These authors emphasize learning-by-doing in working from home and assume
that people have a fatalism bias about the probability of being infected in the
future. Other differences between our paper and theirs are as follows. First, we
explicitly allow for the probabilistic arrival of vaccines and treatments. Second,
we consider the social cost of starting containment too late or ending it too early.
5182

Third, we study “smart containment” policies that make allocations a function

of whether people are infected, susceptible, or recovered.
Guerrieri et al. (2020) develop a theory of Keynesian supply shocks that
trigger changes in aggregate demand that are larger than the shocks themselves.
These authors argue that the economic shocks associated with the COVID-19
epidemic may have this feature. Guerrieri et al. (2020) analyze the efficacy of
various fiscal and monetary policies at dealing with these shocks. In contrast
with Guerrieri et al. (2020), we incorporate an extended version of SIR
dynamics into our model.
Berger et al. (2020) and Stock (2020) study the importance of randomized
testing in estimating the health status of the population and designing optimal

mitigation policies. In contrast with these authors, we explicitly model the
two-way interaction between infection rates and economic activity.
A growing body of work studies the effects of the COVID-19 epidemic
in models in which people differ in their health status as well as along other
dimensions. For example, Glover et al. (2020) study optimal mitigation policies
in a model that takes into account the age distribution of the population. Kaplan,
Moll and Violante (2020) do so in a heterogeneous agent new-Keynesian model.
Faria-e-Castro (2021) studies the effect of an epidemic, modeled as a large
negative shock to the utility of consumption of contact-intensive services, in a
model with borrowers and savers. Buera et al. (2020) study the impact of an
unanticipated lockdown shock in a heterogeneous-agent model.
8. Conclusion
We extend the canonical epidemiology model to study the interaction between
economic decisions and epidemics. In our model, the epidemic generates both
supply and demand effects on economic activity. These effects work in tandem
to generate a large, persistent recession.
We abstract from many important real-world complications to highlight the
basic economic forces at work during an epidemic. The central message of
our analysis should be robust to allowing for those complications: there is an
inevitable trade-off between the severity of the short-run recession caused by
the epidemic and the health consequences of that epidemic. Dealing with this
trade-off is a key challenge confronting policy makers.
Our model also abstracts from various forces that might affect the long-
run performance of the economy. These forces include bankruptcy costs,
unemployment hysteresis effects, and the destruction of supply chains. It is
important to embody these forces in macroeconomic models of epidemics and
study their positive and normative implications.
5183

Appendix. Computing the Equilibrium

This appendix summarizes the equations used to compute the competitive equilibrium of the
benchmark SIR-macro model. The basic SIR-macro corresponds to a particular case in which δc
= δv = κ = 0.
−1
For a given sequence of containment rates, {μt }H
t=0 , for some large horizon, H , guess sequences
s i r H −1
for nt ,nt ,nt t=0 . In practice, we solve the model for H = 250 weeks. Compute the sequence of
the remaining unknown variables in each of the following equilibrium equations:
θ nrt = Aλrbt ,
r −1
ct = (1+μt )λrbt ,
θ r 2
urt = lnctr − n .

2 t
Iterate backward from the post-epidemic steady-state values of Utr :
Utr = u(ctr ,nrt )+βUt+1

r
.
Calculate the sequence for remaining unknowns in the following equations:
(1+μt )ctr = Anrt +t ,
θ nit = φ i Aλibt ,
i −1 i
ct = λbt ,
θ i 2
uit = lncti − n ,
2 t
(1+μt )cts = Anst +t ,
θ s 2
ust = lncts − n .
2 t
Given initial values for Pop0 , S0 , I0 , R0 , and D0 , iterate forward using the following seven equations
for t = 0,..,H −1:
Tt = π1 (St cts ) It cti +π2 (St nst ) It nit +π3 St It ,
Popt+1 = Popt −πdt It ,
St+1 = St −Tt ,
It+1 = It +Tt −(πr +πdt )It ,
Rt+1 = Rt +πr It ,
πdt = πd +κIt2 ,
Dt+1 = Dt +πdt It .
Iterate backward from the post-epidemic steady-state values of Uts and Uti :

Uti = u(cti ,nit )+(1−δc ) (1−πr −πdt )βUt+1
i r
+πr βUt+1 r
+βδc Ut+1 .
Tt
τt = ,
St
Uts = u(cts ,nst )+(1−δv )(1−τt )βUt+1

s r
+δv (1−τt )βUt+1 i
+τt βUt+1 .
5184

Calculate the sequence of the remaining unknowns in the following equations:

i
β Ut+1 −Ut+1
s
−λτ t = 0,
s −1
ct −λsbt (1+μt )+λτ t π1 It Cti = 0.
H −1
Finally, use a gradient-based method to adjust the guesses nst ,nit ,nrt t=0 so that the following
three equations hold with arbitrary precision:
(1+μt )cti = φ i Anit +t ,

μt St cts +It cti +Rt ctr = t (St +It +Rt ),

−θ nst +Aλsbt +λτ t π2 It nit = 0.

To solve the Ramsey problem, we compute the sequence of optimal containment rates, μt , by
maximizing the social welfare function, (14), using the Matlab routine fmincon.m. We explored
the robustness of the numerical results by using Matlab’s global optimization toolbox, including
fminunc.m, patternsearch.m, ga.m, particleswarm.m, surrogateopt.m, and globalsearch.m. None
of these routines finds an optimum that is superior to the one based on fmincon.m.
References
Adda, J. 2016. Economic activity and the spread of viral diseases: Evidence from high frequency data. Quarterly
Journal of Economics 131:891–941.
Alvarez, F., D. Argente, and F. Lippi. 2021. A simple planning problem for COVID-19 lockdown, testing, and
tracing. American Economic Review: Insights. Advance Access published 10.1257/aeri.20200201.
Atkeson, A. 2020. What will be the economic impact of COVID-19 in the US? Rough estimates of disease
scenarios. Working Paper No. w26867, National Bureau of Economic Research.
———. 2021. A parsimonious behavioral SEIR model of the 2020 COVID epidemic in the United States and
the United Kingdom. Working Paper No. w28434, National Bureau of Economic Research.
Atkeson, A., K. Kopecky and T. Zha. 2020. Behavior and the Transmission of COVID-19. Working Paper,
University of California, Los Angeles.
Bennhold, K., and M. Eddy. 2020 Merkel gives Germans a hard truth about the coronavirus. New York Times,
March 11.
Berger, D., K. Herkenhoff, C. Huang, and S. Mongey. 2020. Testing and reopening in an SEIR model. Review of
Economic Dynamics. Advance Access published November 27, 2020, 10.1016/j.red.2020.11.003.
Bootsma, M., and N. Ferguson. 2007. The effect of public health measures on the 1918 Influenza pandemic in
US cities. Proceedings of the National Academy of Sciences 104:7588–93.
Bernanke, B., and J. Yellen (2020) The Federal Reserve must reduce long-term damage from coronavirus.
Financial Times, March 18.
Breban, R., R. Vardavas, and S. Blower. 2007. Theory versus data: How to calculate R0? PLoS One 2:e282.
Buera, F., R. Fattal-Jaef, P. Neumeyer, and Yongseok Shin. 2020. The economic ripple effects of COVID-19.
Working Paper, World Bank.
Chetty, R., J. Friedman, N- Hendren, M. Stepner, and the Opportunity Insights Team. 2020. How did COVID-19
and stabilization policies affect spending and employment? A new real-time economic tracker based on private
sector data. Working Paper, Harvard University.
de Walque, D., J. Friedman and A. Mattoo. 2020. Two tests for a COVID-19 world. Working Paper, World Bank.
5185

Eichenbaum, M. S., S. Rebelo, and M. Trabandt. 2020a. The macroeconomics of testing and quarantining.
Working Paper No. w27104 National Bureau of Economic Research.
———. 2020b. Epidemics in the neoclassical and new Keynesian models. Working Paper No. w27430. National
Bureau of Economic Research.
Eichenbaum, M. S., M. Godinho de Matos, F. Lima, S. Rebelo, and M. Trabandt. 2020. How do people respond
to small probability events with large, negative consequences? Working Paper No. w27988, National Bureau of
Economic Research.
Farboodi, M., G. Jarosch and R. Shimer, 2020. Internal and external effects of social distancing in a pandemic.
Working Paper, University of Chicago, Becker Friedman Institute for Economics.
Farhi, E. and I. Werning. 2020. Dealing with the trilemma: Optimal capital controls with fixed exchange rates.
Working Paper, Harvard University.
Faria-e-Castro, M. Forthcoming. Fiscal policy during a pandemic. Journal of Economic Dynamics and Control.

Ferguson, N., D. Cummings, C. Fraser, J. Cajka, P. Cooley, and D. Burke. 2006. Strategies for mitigating an
influenza pandemic. Nature 442:448—52.
Ferguson, N., D. Laydon, G. Nedjati-Gilani, N. Imai, K. Ainslie, M. Baguelin, S. Bhatia, A. Boonyasiri, Z.

Cucunubá, G. Cuomo-Dannenburg, A. Dighe, I. Dorigatti, H. Fu, K. Gaythorpe, W. Green, A. Hamlet, W.
Hinsley, L. Okell, S. van Elsland, H. Thompson, R. Verity, E. Volz, H. Wang, Y. Wang, P. Walker, C. Walters, P.
Winskill, C. Whittaker, C. Donnelly, S. Riley, and A. Ghani. 2020. Impact of non-pharmaceutical interventions
(NPIs) to reduce COVID-19 mortality and healthcare demand. Working Paper, Imperial College.
Garibaldi, P., E. Moen, and C. Pissarides. 2020. Modeling contacts and transitions in the SIR epidemics model.
Working Paper, London School of Economics.
Ginsberg, J., M. Mohebbi, R. Patel, L. Brammer, M. Smolinski, and L. Brilliant. 2008. Detecting Influenza
epidemics using search engine query data. Working Paper, Centers for Disease Control and Prevention.
Gourinchas, P. 2020. Flattening the pandemic and recession curves. Working Paper, University of California,
Berkeley.
Glover, A., J. Heathcote, D. Krueger, and J. Ríos-Rull. 2020. Health versus wealth: On the distribution effects
of controlling a pandemic. Working Paper, University of Pennsylvania.
Goolsbee, A., and C. Syverson. 2021. Fear, lockdown, and diversion: Comparing drivers of pandemic economic
decline. Journal of Public Economics 193:10.1016/j.jpubeco.2020.104311.
Greenwood, J., P. Kircher, C. Santos, and M. Tertilt. 2019. An equilibrium model of the African HIV/AIDS
epidemic. Econometrica. 87:1081–113.
Guerra, F., S. Bolotin, G. Lim, J. Heffernan, S. Deeks, Y. Li, and N. Crowcroft. 2017. The basic reproduction
number (R0) of measles: A systematic review. Lancet Infectious Diseases 17:e420–e428.
Guerrieri, V., G. Lorenzoni, L. Straub, and I. Werning. 2020. Macroeconomic implications of COVID-19: Can
negative supply shocks cause demand shortages? Working Paper, University of Chicago.
Hatchett, R., C. Mecher, and M. Lipsitch. Public health interventions and epidemic intensity during the 1918
Influenza pandemic. Proceedings of the National Academy of Sciences 104:7582–7.
Jones, C., T. Philippon, and V. Venkateswaran. 2021. Optimal mitigation policies in a pandemic: Social Distancing
and Working from Home. Review of Financial Studies 34:5188–224.
Kaplan, G., B. Moll, and G. Violante. 2020. The great lockdown and the big stimulus: Tracing the pandemic
possibility frontier for the US. Working Paper, University of Chicago.
Kermack, W., and A. McKendrick. 1927. A contribution to the mathematical theory of epidemics. Proceedings
of the Royal Society of London, series A 115:700–721.
Kremer, M. 1996. Integrating behavioral choice into epidemiological models of AIDS. Quarterly Journal of
Economics 111:549–73.
5186

Krueger, D., H. Uhlig, and T. Xie. 2020. Macroeconomic dynamics and reallocation in an epidemic. Working
Paper, University of Pennsylvania.
Lee, B., S. Brown, P. Cooley, R. Zimmerman, W. Wheaton, S. Zimmer, and J. Grefenstette. 2010. A computer
simulation of employee vaccination to mitigate an influenza epidemic. American Journal of Preventive Medicine
38:247–57.
Manski, C. 2010. Vaccination with partial knowledge of external effectiveness. Proceedings of the National
Academy of Sciences 107:3953–60.
Moran, M. 2016. Guidance on treatment of the economic value of a statistical life in U.S. Department of
Transportation Analyses. 2014 Adjustment. Report, U.S. Department of Transportation.
Perrings, C., C. Castillo-Chavez, G. Chowell, P. Daszak, E. Fenichel, D. Finnoff, R. Horan, A. Kilpatrick, A.

Kinzig, N. Kuminoff, and S. Levin. 2014. Merging economics and epidemiology to improve the prediction and
management of infectious disease. EcoHealth 11:464–75.

Philipson, T. 2000. Economic epidemiology and infections diseases. In Handbook of health economics, eds. A.
Culyer and J. Newhouse, vol. 1, 1761ÃŒÃ”99. Amsterdam, the Netherlands: Elsevier.
Pueyo, T. 2020. Coronavirus: Why you must act now. Politicians, community leaders and business leaders: What
should you do and when? Medium, March 10.
Rachel, L. 2020. An analytical model of Covid-19 lockdowns. Working Paper, Bank of England.
Riou, J., and C. Althaus. 2020. Pattern of early human-to-human transmission of Wuhan. Eurosurveillance
25:10.2807/1560-7917.ES.2020.25.4.2000058.
Salje, H., C. Tran Kiem, N. Lefrancq, N. Courtejoie, P. Bosetti, J. Paireau, and A. Andronico. 2020. Estimating
the burden of SARS-CoV-2 in France. Science 369:208–11.
Stock, J. 2020. Coronavirus data gaps and the policy response to the novel coronavirus. Working Paper, Harvard
University.
Toxvaerd, F. 2020. Equilibrium social distancing. Working Paper, University of Cambridge.
Villas-Boas, S., J. Sears, M. Villas-Boas, and V. Villas-Boas. 2020. Are we staying home to flatten the curve?
Working Paper, University of Berkeley.
Viscusi, W., and J. Aldy. 2003. The value of a statistical life: A critical review of market estimates throughout
the world. Journal of Risk and Uncertainty 27:5–76.
U.S. Environmental Protection Agency. Valuing mortality risk reductions for environmental policy. White Paper.
5187

The Macroeconomics of Epidemics: Martin S. Eichenbaum

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

The Macroeconomics of Epidemics: Martin S. Eichenbaum

Uploaded by

Copyright:

Available Formats

The Macroeconomics of Epidemics

Downloaded from https://academic.oup.com/rfs/article/34/11/5149/6213914 by guest on 10 November 2021

As COVID-19 spreads throughout the world, governments are struggling with

[22:50 7/10/2021 RFS-OP-REVF210044.tex] Page: 5149 5149–5187

Downloaded from https://academic.oup.com/rfs/article/34/11/5149/6213914 by guest on 10 November 2021

1 SIR is an acronym for susceptible, infected, recovered, and removed.

[22:50 7/10/2021 RFS-OP-REVF210044.tex] Page: 5150 5149–5187

Downloaded from https://academic.oup.com/rfs/article/34/11/5149/6213914 by guest on 10 November 2021

[22:50 7/10/2021 RFS-OP-REVF210044.tex] Page: 5151 5149–5187

Downloaded from https://academic.oup.com/rfs/article/34/11/5149/6213914 by guest on 10 November 2021

[22:50 7/10/2021 RFS-OP-REVF210044.tex] Page: 5152 5149–5187

The most general version of our model, discussed in Section 6, incorporates

Downloaded from https://academic.oup.com/rfs/article/34/11/5149/6213914 by guest on 10 November 2021

[22:50 7/10/2021 RFS-OP-REVF210044.tex] Page: 5153 5149–5187

Downloaded from https://academic.oup.com/rfs/article/34/11/5149/6213914 by guest on 10 November 2021

1. The SIR-Macro Model

1.1 The preinfection economy

(1+μt )ct = wt nt +t .

Here, wt denotes the real wage rate, μt is a Pigouvian tax rate on

(1+μt )θnt = ct−1 wt .

[22:50 7/10/2021 RFS-OP-REVF210044.tex] Page: 5154 5149–5187

There is a continuum of competitive representative firms of unit measure

The government’s budget constraint is given by

Downloaded from https://academic.oup.com/rfs/article/34/11/5149/6213914 by guest on 10 November 2021

1.2 The outbreak of an epidemic

[22:50 7/10/2021 RFS-OP-REVF210044.tex] Page: 5155 5149–5187

susceptible and infected people, respectively. The parameter π2 reflects the

Downloaded from https://academic.oup.com/rfs/article/34/11/5149/6213914 by guest on 10 November 2021

[22:50 7/10/2021 RFS-OP-REVF210044.tex] Page: 5156 5149–5187

Total population, Popt+1 , evolves according to

Downloaded from https://academic.oup.com/rfs/article/34/11/5149/6213914 by guest on 10 November 2021

Susceptible people. The lifetime utility of a susceptible person, Uts ,

[22:50 7/10/2021 RFS-OP-REVF210044.tex] Page: 5157 5149–5187

The first-order condition for τt is

Infected people. The lifetime utility of an infected person, Uti , is

u1 (cti ,nit ) = λibt (1+μt ),

Downloaded from https://academic.oup.com/rfs/article/34/11/5149/6213914 by guest on 10 November 2021

Recovered people. The lifetime utility of a recovered person, Utr , is

u1 (ctr ,nrt ) = λrbt (1+μt ),

u2 (ctr ,nrt ) = −wt λrbt ,

Government budget constraint. The government budget constraint

Equilibrium. In equilibrium, each person solves his or her maximization

St Cts +It Cti +Rt Ctr = ANt ,

St Nts +It Nti φ i +Rt Ntr = Nt .

[22:50 7/10/2021 RFS-OP-REVF210044.tex] Page: 5158 5149–5187

2. Medical Preparedness, Treatments, and Vaccination

2.1 The medical preparedness model

Downloaded from https://academic.oup.com/rfs/article/34/11/5149/6213914 by guest on 10 November 2021

2.2 The treatment model

[22:50 7/10/2021 RFS-OP-REVF210044.tex] Page: 5159 5149–5187

2.3 The vaccination model

Downloaded from https://academic.oup.com/rfs/article/34/11/5149/6213914 by guest on 10 November 2021

Uts = u(cts ,nst )+(1−δv )(1−τt )βUt+1

[22:50 7/10/2021 RFS-OP-REVF210044.tex] Page: 5160 5149–5187

3.1 Parameter values

Downloaded from https://academic.oup.com/rfs/article/34/11/5149/6213914 by guest on 10 November 2021

[22:50 7/10/2021 RFS-OP-REVF210044.tex] Page: 5161 5149–5187

to the U.S. Department of Agriculture).10 These considerations imply that the

Downloaded from https://academic.oup.com/rfs/article/34/11/5149/6213914 by guest on 10 November 2021

[22:50 7/10/2021 RFS-OP-REVF210044.tex] Page: 5162 5149–5187

Downloaded from https://academic.oup.com/rfs/article/34/11/5149/6213914 by guest on 10 November 2021

The model’s basic reproduction number A statistic widely used to

[22:50 7/10/2021 RFS-OP-REVF210044.tex] Page: 5163 5149–5187